Who Pays for Data Center Overruns? New Insights into Cost-Sharing Models
The rush to build data centers for artificial intelligence is facing a significant hurdle: ballooning construction costs. A detailed industry report, drawing from recent corporate earnings calls, sheds light on how these multi-million dollar overruns are being allocated between infrastructure providers and their tenants.
Dissecting a Major Cost-Sharing Agreement
The report zooms in on a revised lease agreement for a large-scale project. Under the new terms, the operator agreed to absorb the first layer of cost overruns above the original budget, a sum reaching $359.3 million. Any expenses beyond that threshold will be split, with the tenant reimbursing half through additional rental payments over the 20-year lease term, coupled with an agreed-upon return for the operator.
Management pinpointed the root causes, stating that equipment price hikes were not the primary driver. Instead, they cited signing contracts before designs were finalized, leading to a high volume of change orders. Lower onsite productivity and delays in equipment delivery further exacerbated costs. Learning from this, the company plans to sign future leases only after designs are 30% to 60% complete.
Industry Strategies: Fixed Prices and Shifting Capital
Other firms are adopting different risk-mitigation approaches. One company noted that approximately 70% of the budget for its new facility will be covered by Guaranteed Maximum Price contracts, aiming to control core construction costs. It maintained its capital expenditure estimate of $11 to $12 million per megawatt of IT capacity.
The financial landscape is also shifting. The company's project model now assumes debt will finance 80% of costs, down from 90%, dramatically increasing the need for cash equity from roughly $800 million to about $1.5 billion. This highlights the growing impact of financing costs on project economics.
Billions in Commitments Signal Rapid Expansion
The scale of the industry's build-out is further illustrated by another firm's disclosure of $4.4 billion in outstanding purchase and construction commitments, most of which are expected to be incurred within the next 24 months.
These developments underscore that in the AI infrastructure race, robust capital planning, clear risk allocation, and disciplined project execution are paramount. The evolution of financial and contractual models between builders and compute tenants will be crucial for the sector's sustainable growth.